What it is
Raft makes a cluster of servers agree on one ordered log, even when servers crash and messages are lost. One server is elected leader for a term, appends client requests to its log, and replicates them; an entry is committed once a majority stores it. When the leader disappears, followers time out at random moments, so usually one candidate collects a majority of votes first. This simulator runs that protocol on a simulated network, lets you inject crashes, partitions and message loss, and checks the five safety properties of the paper after every event.
When to use it
Use it to see why a leader cut off in a minority cannot commit and loses its entries, why randomised election timeouts avoid split votes, what cluster size buys and costs, and why a new leader appends a no-op. Rewind the timeline to the moment before a failure, change what happens, and compare cluster sizes and timer settings on the same faults. Switch a rule off, such as the log check when voting, and watch the safety panel catch the lost entry; the paper's Figure 8 is a ready-made script.
Gotchas & limits
This is a teaching model of Raft as published: no membership changes, snapshots, read paths or disk persistence, and a crashed node keeps its term, vote and log. Correctness is checked against the five safety properties of the paper, not by comparing with etcd or any other implementation. Timer and network values are illustrative; results show the mechanism, not a prediction for your cluster.